Mindset v3 / Features / Build AI experiences
View as MarkdownAgent Memory
Agent Memory helps your AI agents remember important details about the people they work with, transforming transactional interactions into intelligent, evolving relationships.
NOTE: Agent Memory is currently in Beta, and some features aren't available yet, but are clearly marked with a 'Coming Soon' label.
What is Agent Memory?
Agent Memory helps your AI agents remember important details about the people they work with. Instead of starting every conversation from scratch, your AI can remember that Sarah likes quick updates, John's team is preparing for the Q4 launch, or that the marketing team prefers longer explanations.
Think of it as giving your AI agents the ability to build relationships and provide more personalized help over time.
Your AI agents can create genuine connections by remembering context and personalizing interactions.
No more answering the same questions every time you talk to your AI agent.
Get more relevant, helpful responses that save you time.
Built-in protection for data privacy that meets global standards.
How Agent Memory Works: Three Levels of Control
Agent Memory uses a three-layer system that gives you complete control over how your organization uses AI memory.
Global Agent Configuration - Set your company-wide memory policy
Master Switch: Enable or disable memory collection across your entire organization
Policy Guidance: Define what types of information align with your business goals. For example, "Do not store address information".
Consistency: Ensure all agents follow the same organizational principles
Memory Tab - Turn on memory for your agent
Read Shared Memory: Allow agents to read existing facts about users from the shared memory pool. The agent can access previously stored information about humans from conversations with other agents.
Contribute Memory: Enable agents to save new facts to the shared memory pool. Information learned will be saved for other agents to access.
Fact Collection Guidance: Specific instructions for this agent about what additional information it should record as facts when users mention them. For example: "Record user preferences for cummunication styles, working hours, project priorities..."
Example Configurations:
- Customer Support Agent: Enable both read and contribute, focus on technical preferences, and solution history
- Project Management Agent: Enable both read and contribute, emphasize project context and team dynamics
- HR Agent Agent: Read enabled contribute disabled, to avoid storing sensitive information
Human Facts - Every user has final control over their own information. Users can ask agents what information it has stored about them and can request that the agent deletes/alters those facts.
Opt-Out: Admin users can opt out individual users.
Individual users will be able to opt out (Coming Soon)
Data Control: Request corrections or deletions with GDPR compliance
What Gets Remembered
Agent Memory intelligently categorizes information into meaningful groups:
Job roles, team relationships, project involvement
Meeting styles, information formats
Availability, deadlines, collaboration styles
Project priorities, professional development
Administrative Tools
Human Facts Management
Navigate to any individual's Human Facts tab to access:
📊 Facts Table
Central dashboard showing all collected information, sortable and searchable (Coming Soon)
✏️ Edit Capabilities
Modify any fact while maintaining complete audit trails (Coming Soon)
📥 Export Functions
One-click exports in CSV format for compliance reporting
⚡ Bulk Operations
Select multiple facts for batch operations (Coming Soon)
Currently we can delete all of a user's facts via bulk operations
Compliance Features
Automatic Legal Framework Support:
- EU AI Act (2024): 6+ month logging retention and decision transparency
- GDPR: 30-day response automation for all data subject rights (Articles 15-22)
- CCPA/CPRA: 45-day processing with comprehensive deletion capabilities
- US State Laws: Virginia, Colorado, Connecticut, Utah coverage
- PIPEDA: Full accountability and access rights implementation
Built-In Data Rights:
- Right to Access: Instant visibility of all collected facts
- Right to Rectification: Direct editing with full audit trails (Coming Soon)
- Right to Erasure: Complete deletion with verification across all systems
- Right to Portability: Professional data exports with metadata
Real Examples of How Agent Memory Works
Your customer support AI talks to Sarah, who calls monthly about her software setup. Without memory, every call starts over. With memory, the AI agent immediately knows:
- Her technical setup details
- She prefers brief, technical explanations
- Her team's current project
Your project management AI works with multiple teams. Memory helps it remember:
- Who does what on each team
- Project deadlines and dependencies
- Context across different project phases
Your HR AI handles sensitive discussions. You might set it up to:
Read existing context for continuity
Without storing new sensitive information
Not write new memory
To prevent storing confidential details
Follow strict guidance
About appropriate workplace context
Getting Started
Set Your Foundation
Start with your organization's memory policy. Go to Global Agent Configuration and establish guidelines about what information would genuinely help your AI agents provide better service.
Try It Out
Choose 1-2 AI agents for initial testing. Customer support and project management agents often show immediate value. Enable both **_reading _**and contributing with specific guidance about what to remember.
Watch and Improve
Monitor how your AI agents use memory over the first few weeks. Are they collecting useful information? Are users finding conversations more helpful? Use the Human Facts tab to review what's being learned.
Expand
Based on your trial results, add memory to more AI agents. Each type may need different guidance - your sales agents focus on different information than your technical support agents.
Common Questions
How do I know if Agent Memory is working?
Look for these signs: Users stop repeating basic information, AI agents give more relevant responses, and you see fewer "starting from scratch" conversations. The Human Facts tab will show growing, useful information over time. An admin would know that memory is working on a specific agent if they tell the agent a fact in the Preview tab and the 💡 fact status update shows
What's the difference between reading and contributing memory?
Reading means the AI can use existing information about someone. Contributing means the AI can learn and store new information. You might have some AI agents that only read context, and others that focus on learning new information.
How specific should my memory guidance be?
Be as specific as helpful. Instead of "remember user preferences," try "remember preferred communication style (email vs. phone), typical project deadlines, and team collaboration preferences."
What happens when someone asks to see their data?
Go to their Human Facts tab where you can export all facts about them.
How do I handle deletion requests?
From the Human Facts tab, you can delete individual facts or everything at once.
Are there legal risks?
Agent Memory is designed for compliance with major privacy laws. All actions are logged, users control their information, and you have professional tools for handling data requests. Following the built-in processes keeps you compliant.
My AI agents aren't collecting useful information. What's wrong?
Check your memory guidance. Vague instructions lead to irrelevant information. Be specific about what would actually help your AI provide better service. Also ensure "Contribute Memory" is enabled for agents you want to learn new information.
How much information is too much?
Focus on information that genuinely improves service. If your AI remembers someone's coffee preference but you're a software company, that's probably too much. If it remembers project timelines and communication preferences, that's valuable context.